Learning to Estimate Shapley Values with Vision Transformers

Ian Covert, Chanwoo Kim, Su-In Lee

Transformers have become a default architecture in computer vision, but understanding what drives their predictions remains a challenging problem. Current explanation approaches re…